BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:On Loss Function Landscapes of Neural Networks - Max Niroomand\, U
 niversity of Cambridge
DTSTART:20211110T150000Z
DTEND:20211110T153000Z
UID:TALK162418@talks.cam.ac.uk
CONTACT:Lisa Masters
DESCRIPTION:Machine learning is considered to be one of the most popular s
 tatistical methods of our time. Yet\, machine learning models are commonly
  viewed as black-boxes\, where decision- making is opaque and hard to inte
 rpret by humans. This view can be partially attributed to the way that mac
 hine learning is applied in practice. Usually\, some at best locally optim
 al set of weights is found and accepted as the solution to the learning pr
 oblem. Our work challenges this procedure and shows how\, by looking at la
 rge parts of the loss function landscape (LFL) instead of a single minimum
 \, machine learning can be made more transparent and interpretable. Our ai
 m is to better understand why machine learning works so well\, and we beli
 eve this is best done by understanding the surface of the function that is
  optimised during the learning procedure. \nThis talk will introduce machi
 ne learning in general and the current state of our research specifically.
  Firstly\, it will be shown how properties of the LFL can be exploited to 
 increase accuracy and interpretability of machine learning tasks by over 2
 0% when combining the expressive power of multiple minima in a single clas
 sifier. Secondly\, it will be shown how geometric properties of the LFL ca
 n be employed to guide loss function selection in neural networks. Lastly\
 , some of the most interesting and promising future aspects of this field 
 will be outlined.\n
LOCATION:Zoom and Dept of Chemistry\, Wolfson Lecture Theatre
END:VEVENT
END:VCALENDAR